Troubleshooting High CPU Utilization
Available Languages
Table of Contents
Troubleshooting High CPU Utilization
When High CPU Utilization Is a Problem
Normal Conditions with High CPU Utilization
Other Events Causing High CPU Utilization
Network Symptoms Caused by High CPU Utilization
Determining Interrupt Percentage
Identifying the Cause as System Process or Network Traffic
System Processes and Network Packets
System Processes and Punted Packets
Identifying Network Packets Received by the CPU
Limiting Network Packets to the CPU
Identifying Packets Punted from the Switch Hardware
Identifying TCAM Utilization Issues
Resolving TCAM Utilization Issues
Commands Used for CPU Troubleshooting
Obtaining Documentation and Submitting a Service Request
CPU Utilization Overview
When the switch has completed the boot process, the CPU has two distinct functions. The first function is run the different processes under IOS to carry out the function for a switch operating in a network. The second is to send/receive packets to/from the switching hardware. The CPU is doing both of these functions simultaneously.
The CPU becomes too busy when either an IOS process consumes too much CPU time or the CPU receives too many packets from the switching hardware. When either of these two CPU consumers requires the CPU resource to the detriment of the other, then the CPU is too busy. For instance the CPU is receiving lots of packets from the hardware because there's a broadcast storm on the network. In this case the CPU is so busy processing all the received packets that the other IOS processes aren't given access to the CPU resource. This is just one example of a possible root cause for high CPU utilization. This debugging section will help you to identify other examples and describe how to take corrective action.
Under normal operating conditions, on a non-stackable switch at a minimum, the CPU will have a certain baseline utilization. Depending on the model and the type being used this can range from 5 to 40% If the switch is stacked then, at a minimum, the CPU will operate normally a few percent higher. The number of members in the stack makes a difference on overall CPU utilization. In the stacked switch, the CPU utilization is measured on the master switch only. If the CPU is busy 5% of the time then the CPU is idle the other 95% of the time. The switch will never report CPU utilization at 0%. There are multiple background IOS processes running on timers that execute multiple times a second. This is why even in the simplest of deployments, the switch never reports 0% CPU utilization.
One of the reasons that different ranges and models within those ranges will differ in the baseline utilization is differences in design. Where the earlier models of the switches with little usage of microcontrollers, the later ones do utilize these more. As more tasks are being offloaded to those microcontrollers there is an increase in communication between the CPU and the microcontrollers. The processes this will be reported under are the HULC led and the Redearth Tx an Rx processes.
To determine switch CPU utilization, enter the show processes cpu sorted privileged EXEC command. The output shows how busy the CPU has been in the past 5 seconds, the past 1 minute, and the past 5 minutes. The output also shows the utilization percentage that each system process has used in these periods.
CPU utilization for five seconds: 5%/0%; one minute: 6%; five minutes: 5%In this output, the CPU utilization for the last 5 seconds shows two numbers (
5%/0%).
- The first number,
5%, tells how busy the CPU was in the past 5 seconds. This number is the total CPU utilization for all the active system processes, including the percentage of time at the interrupt level.- The second number,
0%, shows the percentage of time at the interrupt level in the past 5 seconds. The interrupt percentage is the CPU time spent receiving packets from the switch hardware. The percentage of time at interrupt level is always less than or equal to the total CPU utilization.Two other important numbers are shown on the same output line: the average utilization for the last 1 minute (6 percent in this example) and the average utilization for past 5 minutes (5 percent in this example). These values are typical for a nonstacked switch in a small and stable environment.
There can be hundreds of active system processes on the CPU at any time. This number can vary, based on the switch model, the Cisco IOS release, the feature set, and (if applicable) the number of switches in a switch stack. For example, on a stack of Catalyst 3750 switches running the IP base image, there are typically 475 active system processes. The Catalyst 2960 switch running the LAN base image has a smaller number of active processes than a stack of Catalyst 3750 switches. In general, the more features in the Cisco IOS image, the more system processes.
When High CPU Utilization Is a Problem
These sections tell how to identify high CPU utilization and determine if it is a problem:
- Normal Conditions with High CPU Utilization
- Network Symptoms Caused by High CPU Utilization
- Determining Interrupt Percentage
In some instances, high CPU utilization is normal and does not cause network problems. High CPU utilization becomes a problem when the switch fails to perform as expected.
Enter the show processes cpu history privileged EXEC command to see the CPU utilization for the last 60 seconds, 60 minutes, and 72 hours. The command output provides graphical views of how busy the CPU has been. You can see if the CPU has been constantly busy or if utilization has been spiking.
In this example, the CPU spiked to 100 percent 46 hours ago, shortly after the switch was rebooted. It spiked to 87 percent within the past hour.
CPU utilization spikes caused by a known network event or activity are not problems. Even an 87-percent spike might be acceptable, depending on the cause. For example, an acceptable spike could be caused by the network administrator entering a write memory privileged EXEC command on the CLI. A spike is also a normal reaction to a topology change in a large Layer 2 network. See the “Normal Conditions with High CPU Utilization” section for a list of events and activities that cause the CPU utilization to spike.
Over time, the switch operates within a certain sustained CPU utilization range, which is considered the normal operations baseline. You can use the output of the show processes cpu history privileged EXEC command to determine normal operating baseline utilization for the previous 72 hours. CPU utilization exceeding the normal operating baseline for an unknown reason indicates that there is a problem.
In the previous example, the CPU utilization was between 40 and 56 percent for the past 37 hours. We consider anything below 50 percent CPU utilization to be acceptable. A sustained level of approximately 50 percent, such as this example, is also acceptable. A sustained CPU utilization of over 50 percent is potentially problematic. While the switch might appear to operate fine at this CPU utilization level, when the switch CPU sustains a 50 percent utilization, its ability to react to dynamic network events is compromised.
Frequent unexplained spikes to the established normal operating baseline or sudden utilization jumps with no explanations are causes for concern. See the “Determining the Root Cause” section to identify the source of a high CPU utilization problem.
Normal Conditions with High CPU Utilization
In some network deployments, a busy CPU is normal. In general, the larger the Layer 2 or Layer 3 network, the greater the demand on the CPU to process network-related traffic. These are examples of operations that have the potential to cause high CPU utilization:
Spanning Tree
A Layer 2 spanning-tree instance runs for every VLAN configured on a Layer 2 switch by the per-VLAN spanning-tree (PVST) feature. The CPU time utilized by spanning tree varies depending upon the number of spanning-tree instances and the number of active interfaces. The more instances and the more active interfaces, the greater the CPU utilization.
IP Routing Table Updates
When a Layer 3 switch enabled for IP routing receives a large routing table, the switch must process the routing information updates. The CPU utilization spike during processing depends on these factors:
Cisco IOS Commands
Some Cisco IOS commands cause a spike in the CPU utilization. These include:
- The show tech-support privileged EXEC command.
- The write memory privileged EXEC command (particularly if the switch is in a stack).
- The show-running configuration privileged EXEC command on a switch stack master.
- The debug privileged EXEC command to enable debugging of a feature. Printing debug messages to the console increases the CPU utilization as long as debugging is enabled.
Other Events Causing High CPU Utilization
- Frequent or large number of IGMP requests—the CPU processes IGMP messages .
- Large number of IP SLAs monitoring sessions—the CPU generates ICMP or traceroute packets.
- SNMP polling activities, particularly MIB walk—the Cisco IOS SNMP engine executes SNMP requests.
- Large number of simultaneous DHCP requests, such as when links are restored to numerous clients (when the switch is acting as DHCP server).
- ARP broadcast storms.
- Ethernet broadcast storms.
Network Symptoms Caused by High CPU Utilization
A CPU that is too busy affects the ability of the system processes to execute as expected. When the system processes do not execute, the switch (or the directly connected networking devices) react as if there was a network problem. For Layer 2 networks, a spanning-tree reconvergence could occur. In a Layer 3 network, a routing topology could change.
Known symptoms that can occur when the switch CPU is too busy:
- Spanning-tree topology change—When a Layer 2 network device does not receive timely spanning-tree BPDUs on its root port, it considers the Layer 2 path to the root switch as down, and the device tries to find a new path. Spanning tree reconverges in the Layer 2 network.
- Routing topology change, such as BGP route flapping or OSPF route flapping.
- EtherChannel links bounce—When the network device at the other end of the EtherChannel does not receive the protocol packets required to maintain the EtherChannel link, this might bring down the link.
- The switch fails to respond to normal management requests:
– Telnet or SSH sessions that are slow or cannot be started
- UDLD flapping—The switch relies on keepalives from its peer in aggressive mode.
- IP SLAs failures due to SLAs responses beyond the acceptable threshold.
- DHCP or IEEE 802.1x failures if the switch cannot forward or respond to requests.
- Dropped packets or increased latency for those packets routed in software.
- HSRP flapping.
Determining Interrupt Percentage
The CPU utilization history shows only the total CPU utilization over time. It does not show the CPU time spent at interrupt. Knowing the time spent at interrupt is critical for determining the cause of CPU utilization. The CPU utilization history shows when the CPU is consistently receiving network packets, but it does not show the cause.
Enter the Cisco IOS show processes cpu sorted 5sec privileged EXEC command to show the current the CPU utilization and which IOS processes are using the most CPU time. In this example, the CPU is too busy because the sustained utilization is over the baseline of 50%.
The arrow is pointing to the interrupt percentage value. It is the second number in the 5 second utilization percentage.
Its normal for the interrupt percentage to be greater than 0 percent and less than 5 percent. It is acceptable for the interrupt percentage to be between 5 percent and 10 percent. An interrupt percentage over 10 percent should be investigated. See the “Analyzing Network Traffic” section for investigation information.
Determining the Root Cause
When you suspect the CPU is too busy, first determine if it is busy because a system process is taking too much CPU time or because it is receiving too many network packets. The debugging techniques are different for these two root causes. These sections tell you how to identify and troubleshoot the cause:
- Identifying the Cause as System Process or Network Traffic
- Analyzing Network Traffic
- Debugging Active Processes
Note Always refer to the release notes for the specific platform and software release of your switch for any known Cisco IOS bugs. You can eliminate these issues from your troubleshooting steps.
When the switch CPU is busy, management tools such as Telnet or SSH are usually not very useful. We recommend that you use the switch console for debugging CPU utilization issues.
Identifying the Cause as System Process or Network Traffic
To determine how busy the CPU is and which operating system processes are using the most CPU time, enter the show processes cpu sorted 5sec privileged EXEC command. In the output, for
CPU utilization for five seconds,the second number is the interrupt percentage. Use the interrupt percentage to determine if the problem is caused by a system process or high network traffic.If the interrupt percentage is too high relative to the total CPU utilization percentage, the CPU utilization problem is caused by receiving too many packets from system hardware. See the “Determining Interrupt Percentage” section for how to identify a high interrupt percentage.
- A high interrupt percentage indicates too much network traffic. This is the most common cause of high CPU utilization. To troubleshoot, see the “Analyzing Network Traffic” section.
- High CPU utilization percentage with a low interrupt percentage indicates a problem with an operating system process. To troubleshoot, see the “Debugging Active Processes” section.
- When both percentages are high or if you cannot determine whether or not the interrupt percentage is a significant contributor to CPU utilization, first see the “Analyzing Network Traffic” section. If the information provided in this section does not resolve the high CPU utilization problem, see the “Debugging Active Processes” section.
In this example, the CPU utilization is 64 percent, and the interrupt percentage is 19 percent, which is high.The utilization problem is caused by the CPU processing too many packets received from the network. In this case, see the “Analyzing Network Traffic” section.
In the next example, the interrupt percentage is low compared to the CPU utilization percentage (5 percent compared to 82 percent). A high CPU utilization and relatively low interrupt percentage indicates that one or more system processes is taking too much time. In this case, see the “Debugging Active Processes” section.
Analyzing Network Traffic
When the interrupt percentage is high, the root cause of the problem is that the CPU is receiving too many packets. To resolve the problem, you need to find the source of the packets, and either stop the flow, or modify the switch configuration. See these sections:
System Processes and Network Packets
You can identify the type of network packets that are flooding the CPU by the system processes that the switch uses to manage the packets. When the CPU interrupt percentage is high compared to the overall CPU utilization percentage, enter the show processes cpu sorted 5sec privileged EXEC c ommand to determine the most active system process. The CPU can receive multiple packet types and have multiple active system processes. The output lists the most active processes first. The most active system processes are probably responding to the receipt of network packets.
Table 1 lists some common system processes and the associated packet types. If one of the listed system processes is the most active process in the CPU, it is likely that the corresponding type of network packet is flooding the CPU.
See the “Identifying Network Packets Received by the CPU” section to find the source of the packets and how to troubleshoot.
System Processes and Punted Packets
On a Layer 3 switch, when the IP route is not known, the switch hardware punts (sends) IP packets to the CPU for IP routing. Punted packets are handled at the interrupt level and can cause the CPU to become too busy. If the interrupt percentage shown in the command output is high, but the most active processes are not those shown in Table 1 , or no processes seem active enough to justify the CPU utilization, the high CPU utilization is most likely caused by punted packets, as shown in the example output.
Table 2 lists the most active system processes when the CPU is busy acting on punted IP packets. CPU processing of punted packets is not associated with a listed process.
See the “Identifying Packets Punted from the Switch Hardware” section for troubleshooting procedures for punted packets.
Identifying Network Packets Received by the CPU
These techniques to determine the type of packets being sent to CPU can complement each other or can be used individually.
- The switch hardware counts the packets in each CPU receive queue. You can use this count to determine the type of packets being received. See the “Monitoring Packet Counts for CPU Receive Queues” section.
- You can use the debug privileged EXEC command to print to the console all CPU-received packets. The debug command can debug each receive queue separately. See the “Debugging Packets from the Switch CPU Receive Queues” section.
- The switch counts all IP packets that are sent and received. This information is useful to identify any counts that are particularly high and incrementing rapidly. See the “Monitoring IP Traffic Counts” section.
Monitoring Packet Counts for CPU Receive Queues
If the switch is being flooded by a specific packet type, the hardware places that packet type into the appropriate CPU queue and counts it. Enter the show controllers cpu-interface privileged EXEC command to see the packet count per queue.
The switch also counts the CPU-bound packets that are discarded due to congestion. Each CPU receive queue has a packet-count maximum. When the receive queue maximum is reached, the switch hardware discards packets destined for the congested queue. The switch counts discarded packets for each queue. Increased discard counts for a particular CPU queue mean heavy usage for that queue.
Enter the show platform port-asic stats drop privileged EXEC command to see the CPU receive-queue discard counts and to identify the queue discarding packets. This command is not as useful as the show controllers cpu-interface command because the output shows numbers for the receive queues instead of names, and it shows only the discards. Because the switch hardware sees the CPU receive-queue dropped packets as sent to the supervisor, the dropped packets are called
Supervisor TxQueue Drop Statisticsin the command output.The queue numbers in this output for
Supervisor TxQueue Drop Statisticsare in the same order as the queue names in the show controllers cpu-interface command output. For example,Queue 0in this output corresponds torpcin the previous output;.Queue 15corresponds tocpu heartbeat, and so on.The statistics do not reset. Enter the command multiple times to review active queue discards. The command output also shows other drop statistics, some of which are truncated in the example.
See the “CPU Receive Queues” section for more information about CPU queues.
Debugging Packets from the Switch CPU Receive Queues
The hardware inserts packets received from the network into 16 different queues for the CPU. You can identify packets sent to the CPU by enabling debugging for a queue. Use the output from the show controllers cpu-interface privileged EXEC command to learn which queues to start debugging first. See the “Monitoring Packet Counts for CPU Receive Queues” section.
If the output from the show controllers cpu-interface command does not indicate a good place to start, we recommend that you enable debugging on one queue at a time until the console is flooded with debug messages. You use a different debug command to turn debugging on and off for each CPU receive queue.
Before you start debugging the receive queues, use this procedure to prevent other applications from writing to the console, to increase the system log buffer (if it is at the default size), and to put a detailed timestamp on the debug messages. When the debugging session is over, the debug messages are in the system log.
Beginning in privileged EXEC mode, follow these steps to prepare for debugging the CPU queue:
Enable system message logging to a local buffer, and set the buffer size to 12800 bytes.
Configure the system to apply a timestamp to debugging messages or system logging messages.
Be ready to enter the undebug all privileged EXEC command to stop any packet flooding on the console. Even though you do not see a command prompt, the switch accepts the undebug all command at any time. After you enter the command, allow some time for the buffered debug messages to subside and for the debugging message buffer to empty.
The debug messages can help to determine the source of the packet flood. This is useful if the packets are all the same type or originating from the same source.
This is an example of turning on the CPU queues one at a time until console floods.
The sequence of actions in the examples:
- After the debug platform cpu-queue host-q command was entered, a single packet was received. This is normal.
- When the next command, debug platform cpu-queue icmp-q , was entered, the flood began. All packets received on icmp-q are the same. Only three packets are shown. Thus, the CPU is receiving an ICMP packet flood.
- Examine the output for information about the source, including the VLAN (200) and source MAC address (0000.0300.0101) of this packet (shown in bold text).
*Mar 2 22:48:16.947: ICMP-Q:Dropped Throttle timer not awake: Remote Port Blocked L3If:Vlan200 L2If:GigabitEthernet1/0/3 DI:0xB4, LT:7, Vlan:200 SrcGPN:3, SrcGID:3, ACLLogIdx:0x0, MacDA:001d.46be.7541, MacSA: 0000.0300.0101 IP_SA:10.10.200.1 IP_DA:10.10.200.5 IP_Proto:1
- Enter the show mac address-table privileged EXEC command for the VLAN to see the MAC address table and to find the interface where this MAC address was learned. The output shows the packets are being received on interface Gigabit Ethernet 1/0/3 (shown in bold text).
You can use these steps for the different packet types when the CPU is being flooded by a single flow. Continue to enable the debugging of the different CPU queues until the console is flooded. See the “CPU Receive Queues” section for details about the CPU queues.
Beginning in privileged EXEC mode, follow these steps to view the system log with the debug messages:
Set the number of lines on the terminal screen for the current session to 0.
Set the terminal length to 30, or reset back to the original value.
Note If you modified the configuration before debugging by increasing the system log buffer or adding a timestamp, consider returning these settings to the default configuration when debugging is complete.
Monitoring IP Traffic Counts
The switch counts all IP packet types received by the CPU. These packet counts do not include the IP packets switched or routed in hardware. Enter the show ip traffic privileged EXEC command to display the IP packet type counts. The output breaks down the IP packets into Layer 4 types (that is, ICMP, multicast, ICMP, or ARP). When a particular count is incrementing rapidly, the IP packet type is probably flooding the CPU.
Limiting Network Packets to the CPU
You can prevent problem network packets from impacting the CPU utilization by stopping them at the ingress interface:
- To limit Ethernet broadcast or multicast packet storms, use the storm-control { broadcast | multicast | unicast } level { level [ level-low ] | bps bps [ bps-low ] | pps pps [ pps-low ]} interface level configuration command. See the “Configuring Port-Based Traffic Control” chapter in the switch software configuration guide.
- If the root cause of the high CPU utilization is a Layer 2 loop, the spanning tree configuration could be the problem. See the “Configuring STP” chapter in the switch software configuration guide.
- Policing traffic can limit the number of packets that enter a switch. Policing can deny ingress traffic, limit it to a specific bits-per-second rate, or permit some traffic while limiting other traffic. You can police traffic on the MAC address, the IPv4 header, the IPv6 header (if IPv6 is supported on the switch), or the Layer 4 port number. See the chapters on “Configuring Network Security with ACLs,” “Configuring IPv6 ACLs” (if supported on the switch), and “Configuring QoS” in the switch software configuration guide.
- To prevent IP ARP packets from affecting the CPU utilization on Layer 3 switches, configure Dynamic ARP Inspection (DAI), and enter the ip arp inspection limit { rate pps [ burst interval seconds ] | none } interface configuration command to use the rate-limiting feature. See the chapter on “Configuring Dynamic ARP Inspection” in the switch software configuration guide.
Identifying Packets Punted from the Switch Hardware
As part of normal Layer 3 switch operation, when the IP route is not programmed into the switch hardware, the hardware punts IP packets to the CPU for IP routing. Punting occasional IP packets to the CPU is normal and expected, but if too many IP packets are punted, the CPU becomes too busy.
The switch counts every IP packet that the hardware punts to the CPU for IP routing. You can use the show controllers cpu privileged EXEC command to see the number of packets put into each CPU receive queue. When the switch hardware is punting packets, the row named
sw forwardingis incrementing. Enter the command several times to see if the counts forsw forwardingare rapidly incrementing.You can also use the show platform ip unicast statistics privileged EXEC to show the same information about punted packets. The punted IP packets are counted as
CPUAdj, shown in bold in this example.These statistics are updated every 2 to 3 seconds. Enter the command multiple times to see the change in the
CPUAdjcounts. When theCPUAdjcounts are rapidly incrementing, many IP packets are being forwarded to the CPU for IP routing.Identifying TCAM Utilization Issues
On a Layer 3 switch, the hardware uses the TCAM to contain the IP routing database. TCAM space for Layer 3 routing information is limited. When this space is full, new routes learned by Cisco IOS cannot be programmed into the TCAM. If IP packets received by the switch hardware have a destination IP address that is not in the TCAM, the hardware punts the IP packets tothe CPU.
To see if the TCAM is full, enter the show platform tcam utilization privileged EXEC command.
In the example, the
IP indirectly-connected routesresource is full even though the output shows only 2047 of 2048 maximum as in use. If the switch TCAM is full, the hardware routes packets only for destination IP addresses that are in the TCAM. All other IP packets that had a TCAM miss are punted to the CPU. A full TCAM and increasingsw forwardingcounts from the show controllers cpu-interface command output means that punted packets are causing high CPU utilization.Cisco IOS learns about routes from routing protocols—such as BGP, RIP, OSPF, EIGRP, and IS-IS—and from statically configured routes. You can enter the show platform ip unicast counts privileged EXEC command to see how many of these routes were not properly programmed into the TCAM.
This output shows 693 failures. You can use this statistic to determine how many additional TCAM resources are needed to hold the routes being advertised in the network at this time.
To view the number of the number of route entries used by each routing protocol, enter the show ip route summary privileged EXEC command.
Changing the SDM Template
The switch database management (SDM) template allocates the limited TCAM resources for different forwarding types. To resolve TCAM utilization issues, you should choose the appropriate SDM template for the switch application.
Enter the show sdm prefer privileged EXEC command to see the active SDM template on a switch:
The default template on this switch allows only 2048 indirect Layer 3 routes in the TCAM. To allocate more TCAM resources to indirect Layer 3 routes, you must reduce some other TCAM resources. Use the sdm prefer template-name global configuration command to change to a template that reserves more resources for IP routing.
To see a list of available SDM templates for your switch, enter the show sdm templates all privileged EXEC command.
Note See the chapter on “Configuring SDM Templates” in the switch software configuration guide to see the templates available on your switch and the reserved TCAM resources for each template.
Optimizing IP Routes
When it is not possible or practical to change the SDM template on a Layer 3 switch, you can reduce the number of routes in the TCAM by using summary routes or by filtering routes.
Using summary routes reduces the routing table size. You enable summary routes on peer routers. Route summary is enabled by default for RIP and EIGRP and disabled by default for OSPF. To learn about route summary, see the “Configuring IP Unicast Routing” chapter in the software configuration guide (only Layer 3 switches).
You can use route filtering to prevent unwanted routes from being programmed into the TCAM. For information about OSPF route filtering, see the feature guide at this URL: http://www.cisco.com/en/US/docs/ios/12_0s/feature/guide/routmap.html
Debugging Active Processes
If the CPU utilization percentage is high and the interrupt percentage is low, the high CPU utilization is caused by one or more system processes consuming the CPU resource. This is less common than high CPU utilization caused by the receipt of network packets. When a system process is consuming most of the CPU resources, an event usually triggered the process to become active. Review the syslog for any unusual events.
Table 3 lists some processes that consume CPU resources with normal and high percentages, possible root cause for the process, and suggested actions.
Review the syslog for a physical link losing and gaining connectivity.
Too many ACLs configured on the switch in a short time period. This can occur when the ACLs are being applied in an automated fashion (from a script).
See the “SNMP Engine Process” section.
SNMP Engine Process
The SNMP engine system process is only active when the switch is receiving SNMP queries. The required CPU time is directly proportional to the number of SNMP query packets received. Each received SNMP query packet is processed at the interrupt level before it is forwarded to the SNMP engine system process. When the SNMP engine process is busy, the interrupt percentage shown in the output of the show processes cpu sorted command also shows some non-zero interrupt percentages. The interrupt percentage is minor compared to the percentage of CPU utilized by the SNMP engine system process.
It is important to determine the switch baseline CPU utilization when evaluating CPU utilization for the SNMP engine system process. The switch typically receives SNMP queries at regular intervals, and the SNMP Engine system process consumes CPU resources handling the queries. The intervals are shown in the output for the show processes cpu history command. See the examples in the “When High CPU Utilization Is a Problem” section.
These scenarios can cause heavy CPU usage by the SNMP engine system process:
Helpful Information
CPU Receive Queues
A packet sent to the CPU by switch hardware goes into one of 16 CPU queues, depending on the packet type. Each queue is given a priority, allowing the CPU to drain higher priority queues before lower priority queues. Each queue has some reserved memory in hardware to hold packets for the queue so that one queue or packet type cannot use all the available memory.
These are the CPU queues with their uses:
- rpc—Remote procedure call. Used by Cisco system processes to communicate across the stack.
- stp—Spanning Tree Protocol. A Layer 2 protocol with its own queue and used for protocol packets such as LACP and so on.
- ipc—Interprocess communication. Used by Cisco system processes to communicate across the stack.
- routing protocol—Used for routing protocol packets received by other network devices.
- L2 protocol—Used for protocol packets such as UDLD, and so on.
- remote console—Used for packets when you enter the session switch-number privileged EXEC command on a stack master switch to open the console on another switch member.
- sw forwarding—Used for packets punted by the hardware for the CPU to route.
- host—Used for packets with a destination IP address matching any switch IP address. Also IP broadcast packets.
- broadcast—Receives Layer2 broadcast packets.
- cbt-to-spt—Receives multicast packets for PIM_SM.
- igmp snooping—A queue for IGMP packets.
- icmp—A queue for ICMP redirect packets.
- logging—Used for receive packets generated by hardware for ACL logging.
- rpf fail—A queue for reverse path forwarding failures.
- dstats—drop stats. Not used during normal operation.
- cpu heartbeat—Used by the CPU receiving the packets it sends to itself.
Commands Used for CPU Troubleshooting
Additional Documents
Another document on Cisco.com focuses on specific high utilization issues in the Catalyst 3750 switch, although the information also applies to other switches. See Catalyst 3750 Series Switches High CPU Utilization Troubleshooting.
Unresolved Issues
If the troubleshooting steps in this document do not help you to determine the root cause of high CPU utilization, you should contact the Technical Assistance Center (TAC) for Cisco. The technical assistance engineer will want to see the same information that you have gathered in the debugging efforts. Have this information ready when you contact Cisco technical support to reduce the time to resolve the problem.
Note See the next section for a link for submitting a service request.
Obtaining Documentation and Submitting a Service Request
For information on obtaining documentation, submitting a service request, and gathering additional information, see the monthly What’s New in Cisco Product Documentation , which also lists all new and revised Cisco technical documentation, at:
http://www.cisco.com/en/US/docs/general/whatsnew/whatsnew.html
Subscribe to the What’s New in Cisco Product Documentation as a Really Simple Syndication (RSS) feed and set content to be delivered directly to your desktop using a reader application. The RSS feeds are a free service and Cisco currently supports RSS Version 2.0.
Cisco and the Cisco logo are trademarks or registered trademarks of Cisco and/or its affiliates in the U.S. and other countries. To view a list of Cisco trademarks, go to this URL: www.cisco.com/go/trademarks . Third-party trademarks mentioned are the property of their respective owners. The use of the word partner does not imply a partnership relationship between Cisco and any other company. (1110R)
Any Internet Protocol (IP) addresses used in this document are not intended to be actual addresses. Any examples, command display output, and figures included in the document are shown for illustrative purposes only. Any use of actual IP addresses in illustrative content is unintentional and coincidental.
FeedbackContact Cisco
- Open a Support Case
- (Requires a Cisco Service Contract)